What is AI share of voice, and how do we compare with competitors?

Your sales director mentions it in passing. Two of the last five prospects had also looked at a firm you barely think of as a competitor, and both said ChatGPT had suggested it. You know exactly who you are up against in pitches. You have no idea who you are up against in AI answers.

That blind spot is getting more expensive. In G2's April 2026 survey of 1,076 software buyers, 51% said they now start their research in an AI chatbot, and 69% had chosen a different vendor from the one they first planned because of what a chatbot told them. 6sense's 2025 Buyer Experience Report found buying groups had ranked their preferred vendors before first contact in 94% of cases. The shortlist forms before anyone calls you, and the lost enquiry never shows up in a report.

Checking your own name in ChatGPT once tells you very little. Being named in one answer in five sounds poor until you learn your nearest rival is named in none, and fine until you learn they are named in every one. AI share of voice is the number that settles it. Here is how to calculate it, find the websites that name your rivals and not you, and pick the gaps to win first.

#What is AI share of voice?

AI share of voice is the proportion of firm mentions in AI answers that go to you, for the questions your buyers ask. Count every firm named across the answers to a fixed set of questions, then divide your mentions by the total.

The idea comes from advertising and PR. Here the conversation is the list of names an assistant gives when a buyer asks who to hire: usually a short list rather than ten links, though the length varies by assistant and question. Tools define the measure in slightly different ways, so check which one you are reading.

Ahrefs defines its AI share of voice as "the percentage of AI responses within your topic set that mention or cite your brand versus competitors", which is closer to a presence rate. Either works if you use the same definition every time, for every firm.

Example: you ask 40 buyer questions and the answers name firms 100 times in total. Your firm is named 15 times, Competitor A 50 times, Competitor B 20 times and Competitor C 15 times. Your share of voice is 15% and Competitor A's is 50%. On its own, 15 mentions sounds like a start. Next to a rival holding half the conversation, it shows which firm the assistants treat as the default.

#How do we calculate our AI share of voice?

Run a fixed set of buyer questions through the assistants your buyers use, record every firm named, and divide. How to choose the questions and run them cleanly is covered in our guide to finding out whether ChatGPT recommends your firm: 20 to 50 questions in your buyers' words, each asked several times a month in a clean session, with a mention counted only when it repeats.

A clean session matters because you are testing what a stranger hears. In ChatGPT, a Temporary Chat set to unpersonalised does not use memory or custom instructions. We would not rely on logged-out sessions alone, as they may not match what a signed-in UK buyer is shown. Include Microsoft Copilot alongside ChatGPT and Google: Forrester found in January 2026 that business buyers are four times as likely as consumers to use it.

The competitor benchmark then adds four things to your normal tracking:

1. Record every firm named, with its order and role: the recommendation, an alternative, or a passing mention.
2. Count mentions and work out each firm's share, with presence and first-mention rates alongside.
3. Split the results by assistant and question type. A rival that leads only on shortlist questions ("who are the best...") has a different advantage from one that leads on the problem questions buyers ask earlier.
4. Copy the cited sources into the same sheet. ChatGPT search, Perplexity, Copilot and Google's AI features show the pages they drew on. They feed the citation gap below.

#Who should we benchmark against?

Benchmark against the three to five rivals your sales team meets most often, plus the firms the assistants actually name for your buyers' questions. The two lists rarely match, and the differences are often the most useful part of the exercise.

Run a handful of buyer questions before you settle the list. You will often find a firm you rarely meet in pitches named again and again, such as a national player or a small specialist with an unusually clear website. If the assistants name them, that is where buyers are meeting them.

#Which tools show competitor mentions in ChatGPT?

Ahrefs Brand Radar and Semrush's AI Visibility Toolkit both report how often AI assistants mention you and your competitors, and which sources they cite. For a first benchmark, a spreadsheet does the same job and forces you to read the answers.

Ahrefs says Brand Radar tracks mentions across AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini, Copilot and Claude, lets you "benchmark your brand against competitors in AI search", reports citations, and accepts custom prompts for "the exact questions your buyers ask". Semrush's guide to its AI Visibility Toolkit (November 2025) shows how to compare your ChatGPT visibility with competitors and lists "domains that cite competitors but not you".

Vendor tools run questions from their own systems, so check a sample by hand before you trust a trend.

#What do the results tell us?

Three things matter: your share against the leader's, the questions where nobody is named consistently, and the assistants where you are missing.

• The gap to the leader. The firm with the largest share is the assistants' default. Take its places question by question, starting where your claim is stronger.
• The questions nobody owns. Where no firm is named consistently, or the assistant names no one, there is no incumbent to displace.
• The assistants that leave you out. Treat this as a question, not a diagnosis. An assistant searching the web live may have found a recent page, one answering from memory reflects older material, and some lean on different sources. Check what each cites for your rivals first.
• What is actually said. A mention that describes you wrongly or uses an old name is a separate, usually quicker, fix. See why ChatGPT describes a firm wrongly.

ID-Pal, the identity verification firm, shows why the comparison matters. It had bought NorthRow, a UK business verification platform, yet "AI could describe ID-Pal. It never recommended it." The case study records the before and after:

It also showed where to work next: "Where a page has been rebuilt, ID-Pal appears. Where it has not yet, it does not, which tells us exactly what to build next."

#Why does ChatGPT recommend my competitor and not us?

Usually because the assistant can find the competitor described consistently in several places it trusts, and cannot find the same for you. The reasons are mostly visible from outside.

• Coverage by others. Muck Rack's May 2026 analysis of more than 25 million links cited by ChatGPT, Claude and Gemini found earned media made up 84% of them and paid content 0.3%.
• Other sites that reference them. An SE Ranking study reported in November 2025 found referring domains were the strongest predictor of ChatGPT citations.
• Reviews and listings. In G2's survey, 45% of software buyers said review-site citations were the most confidence-inspiring signal in an AI answer.
• Pages that answer buyer questions. A clear guide to the exact problem beats a general services page.
• A consistent description across website, LinkedIn, directories and press, so the assistant can name the firm with confidence.

We explain how assistants weigh these sources in why ChatGPT sounds unsure about a firm and whether backlinks still matter for AI search.

#How do we find which websites ChatGPT cites for our competitors?

Export the cited URLs for every answer, reduce them to domains, and list the domains that cite your rivals but not you. The top ten become your PR and listings target list. We call this the citation gap.

#Step 1: Export the cited URLs for every answer

Copy every source URL from your benchmark answers into a sheet, next to the question, the assistant and the firms named. Ahrefs and Semrush export the same data.

#Step 2: Reduce each URL to its domain and note who it names

Open each page and note which firms it mentions: you, each rival, or none.

#Step 3: List the domains that cite rivals but not you

Filter for domains that mention at least one competitor and never mention you, and count how many answers each appears in. Domains that recur across questions and assistants matter most.

#Step 4: Turn the top ten into a target list

For each of the top ten, choose the route in: a directory entry, a review profile, a contributed article or comment in a trade title, an award entry, or a correction to an out-of-date profile. Give each an owner and a date. This is where our Reputation That Wins work usually starts.

#Where are the gaps we can win first?

Start where there is no incumbent, where a rival's lead rests on a thin page, or where a sector has gone unclaimed. In our experience, four gaps come up again and again: questions with no clear answer; topics a competitor covers with a single page, which a more useful set of pages answering the follow-up questions can overtake; sectors nobody has claimed, such as "accountants for dental practices" when rivals are named only for "accountants for small businesses"; and the sources at the top of your citation gap.

Comparison questions deserve a special look, because a competitor often writes the answer by default. Helm Club, a founder membership community, took this route: "We built the structured data AI engines read, then a comparison page for every question a founder asks when weighing up peer groups: Helm against YPO, Vistage and the rest." The case study records: "Every comparison page we have published ranks first for its question on Google and the AI engines. The pages not yet built are the ones where a competitor still writes the answer."

Give each gap a score out of three: how much buyers care, how strong your expertise is, and how hard it would be for a rival to follow. Work on the two or three highest first. The effect builds over time, as we explain in how content, PR and a better website work together.

#What is a good share of voice in AI search?

There is no published benchmark for a good AI share of voice, and a single figure would mislead, because it depends on how many firms compete in your market and how the measure is defined. Compare yourself with the leader on the questions that bring in revenue.

In our view, three readings matter more than the headline: whether you lead on any of your ten most valuable questions, whether the gap to the leader is closing each quarter, and whether your first-mention rate is rising.

#How often should we run the comparison?

Track your own fixed list several times a month, and run the full competitor benchmark with the citation gap every quarter, using the same questions. Our guide to building a simple AI visibility dashboard shows a layout.

#What to do this month

1. Ask your sales team which competitors come up most, add any firm the assistants name unexpectedly, and settle on three to five.
2. Fix a list of 20 to 50 buyer questions in your buyers' words, and choose one share-of-voice definition.
3. Run the list through ChatGPT, Google's AI Overviews and AI Mode, Copilot, Gemini, Perplexity and Claude in clean sessions, recording every firm named and every source cited.
4. Work out each firm's share of voice and presence rate, by assistant.
5. Build your citation-gap list and turn the top ten domains into named PR, listing and review targets.
6. Pick the two or three unowned questions you are best placed to win, and book the next full run for three months' time.

#Frequently asked questions

#How many competitors should we compare ourselves with?

Three to five suits most firms. Fewer and you may miss the firm quietly taking your place; more and the tracking becomes a chore nobody keeps up. Include any firm the assistants name repeatedly, even if you rarely meet it in pitches.

#Why do we get different answers each time we ask?

AI assistants write each answer fresh, and some search the web live, so the firms named vary between runs. That is why a single check tells you little. Asking a fixed set of questions several times and counting only repeat mentions gives a number you can trust.

#What if we are far behind the leading competitor?

That is the normal starting point, and it is recoverable. Assistants that search the web live can reflect a new page or article within weeks, while answers drawn from a model's memory change only when it is retrained, which takes months. Start with the questions nobody owns.

#How do we know the benchmark reflects real buyers?

Check it against what your sales team hears. If the firms leading the AI answers are the ones prospects mention as alternatives, the benchmark is picking up real behaviour. If not, your questions may be written in your language rather than your buyers'.

#Can a competitor pay to be named ahead of us?

Not inside the answer. ChatGPT has shown labelled ads to UK users on its Free and Go plans since 11 August 2026, and OpenAI says the ads do not change the answer itself. Being named in the answer is still earned.

To see where you stand before building your own spreadsheet, book a free Growth Review: 30 minutes with David Adams, with your site, two competitors and what Google and ChatGPT say about all three checked beforehand, and three fixes in writing afterwards.

#Sources